In a double-sided document with a recto and a verso, there is a bleed-through phenomenon of the document so that a part or all of a text on either of the sides may be visible from the other side. This bleed-through phenomenon may impair the readability of the document and automatic character recognition on the document. Consequently, bleed-through pixels in the document have to be processed to improve the readability of the document or the accuracy of automatic character recognition on the document.
The bleed-through pixels can be processed only if accurate locations of the bleed-through pixels are detected, that is, those bleed-through pixels are judged from the pixels in the document images.
Generally, in a conventional solution, an appropriate threshold is selected, and the pixel value of a pixel is compared with the threshold to differentiate bleed-through from a text. Apparently, this solution is rather coarse because there is a low difference between a foreground pixel represented by the text and a bleed-through pixel in the case of serious bleed-through whereas there is an insignificant difference between a background pixel and a bleed-through pixel in the case of slight bleed-through. Consequently, there is a poor effect of processing in the conventional solution.
There is a need of a bleed-through detection method of and bleed-through detection apparatus for detecting accurately the location of a bleed-through pixel in a double-sided image pair including a recto image and a verso image.